Question

I’m trying to port a program from Red Hat Linux to IBM AIX, but the precision on the IBM system seems different to the Red Hat system. Does anyone know why this is happing? Here is an example:

program main
implicit none
real(8) a
a=0.2D0
print *,a
end program

On Red Hat I get output:

0.200000000000000

On AIX I get output:

0.200000000000000011

The version of the system I’m using:

IBM XL Fortran for AIX, V13.1 (5724-X15)
Version: 13.01.0000.0005

and

Linux 2.6.18-8.el5 x86_64

ifort Version 10.0

Answer 1

This is probably just a difference in the default number of digits printed. 0.2 cannot be accurately represented as a binary floating point number, so you are always going to have some garbage in the least significant digits. The default number of digits printed for a float with ifort is simply set to be low enough that you don’t see it in this case. The actual calculation (most likely) has the same precision. Try this with both compilers:

print '(f24.20)',0.2d0

This will show you that ifort also has the ones at the end. On my computer, it prints 0.20000000000000001110.